Muse Glimmer
Meta's 30 billion parameter open-weight model from August 2026, small enough to run on a single consumer graphics card.
Muse Glimmer is the first open model from Meta’s Superintelligence Labs, released in August 2026 under the Apache 2.0 licence. It was distilled from the larger Muse Spark 1.2, meaning a smaller model was trained to imitate a bigger one. At 30 billion parameters, and compressed to roughly 4 bit, it fits in under 20 GB of memory, which puts it within reach of a gaming PC or a well equipped Mac.
It handles text, images and video, and was tuned specifically for agent work, so it can hold a chain of tool calls together rather than only chatting. Hugging Face shipped support in transformers, llama.cpp and vLLM on the day of release. On agent benchmarks it beats larger rivals such as Gemma4-31B, while losing to Qwen3.6-27B on several coding and desktop tests.